HPUX 11.31 ia64 memory usage

[catastro]

Hi all. I am trying to get out of the system the memory utilization data, and I am not able to see it. I already got the kmeminfo tool, and I get that only 9% of memory is free.
In glance I get the same metrics, but I am not able to get a list of processes that are getting all the memory. I know that with glance I can print them, but I am not able to print all the pages (just the first one). Do you know another way to ge this info?

----------------------------------------------------------------------
Physical memory usage summary (in page/byte/percent):

Physical memory = 16743725 63.9g 100%
Free memory = 1558833 5.9g 9%
User processes = 0 0.0b 0% details with -user
System = 158 632.0k 0%
Kernel = 0 0.0b 0% kernel text and data
Dynamic Arenas = 2070068 7.9g 12% details with -arena
misc region are = 219240 856.4m 1%
asyncdsk variab = 211708 827.0m 1%
reg_fixed_arena = 190162 742.8m 1%
spinlock_arena = 172277 673.0m 1%
FCACHE_ARENA = 144862 565.9m 1%
Other arenas = 1131819 4.3g 7% details with -arena
Super page pool = 1359176 5.2g 8% details with -kas
Static Tables = 977005 3.7g 6% details with -static
pfdat = 817564 3.1g 5%
vhpt = 131072 512.0m 1%
text = 8978 35.1m 0% vmunix text section
inode = 7616 29.8m 0%
bss = 6713 26.2m 0% vmunix bss section
Other tables = 5060 19.8m 0% details with -static
Buffer cache = 158 632.0k 0% details with -bufcache

[klb]

ipcs -am ( active shared memory segments )

ps -efl |sort -rn +9 ( SZ of each proc, see ps man page for SZ definition )

Glance should give you a nice summary of memory use including:

Sys Mem
User Mem
Buf Cache
Free Mem

Hth,

-klb

[Vivek_Pendse]

Hi,

When you print the first page in glance, then click "+" to print the next page.
If you click "?" sign, then you can get all the help related to glance.

Thanks,
Vivek

[Laurent Menase]

Hi Catastro,
you can use glance adviser mode like shown in examples at
/opt/perf/examples/adviser

it will give you a direct listing.

[catastro]

Hi. I got the new version of kmeminfo from HP. Now we are seeing the ways to reduce the memory that uses the system.
I have looked at this thread, and I will probably try some of this:

http://h30499.www3.hp.com/t5/System-Administration/HPUX-11-31-the-quot-sys-mem-quot-is-too-high-it-eats-20-of-the/m-p/4749103#M388052

tool: kmeminfo 10.19 - libp4 9.543 - libhpux 1.372 - HP CONFIDENTIAL
unix: /stand/current/vmunix 11.31 64bit Tukwila on host "clase"
core: /dev/kmem live kernel!
link: Thu Apr 28 17:02:10 METDST 2011
boot: Wed May 11 14:27:29 2011
time: Thu May 12 14:42:06 2011
nbpg: 4096 bytes


----------------------------------------------------------------------
Physical memory usage summary (in page/byte/percent):

Physical memory = 4168086 15.9g 100%
Free memory = 266047 1.0g 6%
User processes = 1303104 5.0g 31% details with -user
Detached SHMEM = 7 28.0k 0% details with -shmem
System = 2615567 10.0g 63%
Kernel = 1872944 7.1g 45% kernel text and data
Dynamic Arenas = 1665989 6.4g 40% details with -arena
bc_bpdata_32K_a = 283296 1.1g 7%
misc region are = 133654 522.1m 3%
FCACHE_ARENA = 128741 502.9m 3%
reg_fixed_arena = 124172 485.0m 3%
spinlock_arena = 121190 473.4m 3%
Other arenas = 874936 3.3g 21% details with -arena
Super page pool = 86834 339.2m 2% details with -kas
DMA32 free pool = 65256 254.9m 2%
Emergency pool = 8722 34.1m 0% system critical reserve
UAREA's = 16960 66.2m 0%
Overflow pte's = 32952 128.7m 1%
Static Tables = 263608 1.0g 6% details with -static
pfdat = 203519 795.0m 5%
vhpt = 32768 128.0m 1%
text = 9107 35.6m 0% vmunix text section
inode = 7616 29.8m 0%
bss = 6723 26.3m 0% vmunix bss section
Other tables = 3874 15.1m 0% details with -static
File Cache = 742623 2.8g 18% details with -ufc

For starting, I am thinking in setting vx_ninode to 131072 or less (now it has 0, automatic), set base_page_size to 16 (now it has 4) and set vx_ifree_timelag to -1 (now it has 0).
Any suggestion about it?

[Laurent Menase]

you can see in the kmeminfo output
kmeminfo 10.19 - libp4 9.543 - libhpux 1.372 - HP CONFIDENTIAL
---------------
kmeminfo is a support tool which should be used only on hp request. To have an interpretation of the result, hp support will be the place to go.

[Noé · Accepted solution]

With this command you can see the virtual memory used by each process (sort by the third column in KB)


# export UNIX95=1;ps -ef -o comm,pcpu,vsz,args |sort -n -k 3

Regards.

[catastro]

Wel, I did some digging into performance white papers from HP, and modified some parameters.

I modified the vx_ninode, as it had 0 value (default) and should be in a value of 131000 (aprox) for servers with a memory size of more than 3GB. And modified as well the vxfs_ifree_timelag to -1 in order to stop freeing inodes, as I have many FS.
I also reduced the value of filecache_max and filecache_min as they were not being used.

With this changes, I got 33% more of free memory. Before making these changes I asked HP about them, specially the vxfs_ifree_timelag.
Thanks for all your help.

[catastro]

So, the solution is to tune some kernel parameters. In my case, I changed the filecache_min/max, vx_ninode and vxfs_ifree_timelag parameters values.